Sample composition in online studies
|Convenor||Mr Ulrich Krieger (German Internet Panel, SFB 884, University of Mannheim )|
We investigate what is the gain in terms of representativeness of proposing the equipment to non-Internet units in a web panel using tablets: the ELIPSS panel. We find that the number of non-Internet units that accept to participate is low, because of lower response rates but also due to the limited proportion of non-Internet units in the population. They also participate less in the specific surveys once they are panellists. However, they are very different from the Internet units. Therefore, there are a few important variables on which including them allows obtaining a more representative sample of
In a series of 4 studies, we compared attitudinal measures using opt-in samples with responses from probability samples. In addition, we compared a series of attitudinal measures of presidential approval over 10 years using opt-in samples with aggregate values obtained from probability samples. In all cases, we found significant convergence of results obtained from opt-in non-probability samples with those obtained from probability samples. We discuss both the similarities and differences in results.
Web panels does not usually fulfil the basic demands with regard to representativeness. A simple random sample in a web-panel, e.g. advance stratified, would therefore, even if all participants provide a reply, not be representative, which is usually true of samples selected with a known probability in the overall population. But how close can you get? The discussion paper shows an example from Statistics Denmark, where representative sample surveys have asked for e-mail with regard to contact for new surveys. Proportional sampling by sex, age and geography are compared to proportional sample surveys surveys with more register
The aim of this work is to test whether web-surveys are a reliable tool to collect information about subjective well-being. The Global Entrepreneurship Monitor (GEM) data of 2013 provides information from a sample of 2000 people representative of the Luxembourgian population. Half of the sample answered to telephone interviews, whereas the remaining half used a web-survey. To test whether the use of web-surveys alters people’s self-assessment of their well-being, we use regression, decomposition analysis and propensity score matching. We identify a downward bias of